An overview on automatic design of robot controllers for complex tasks
This is an incremental overview for researchers in robotics and evolutionary computation.
The paper reviews methodologies for automatically designing robot controllers for complex tasks with large abstraction gaps, covering evolutionary setup components and bootstrapping strategies.
In this paper we will explore different available methodologies to automatically design controllers for tasks that spans many level of abstraction, where the gap between primitive behaviours and the task definition is high. A good understanding of your evolutionary setup is needed to choose the correct strategy with which to tackle complex tasks thus we'll first review the most used types of each element composing an evolutionary setup (controllers, objective functions, ect.) then we'll move the focus on the bootstrapping problem and on the different strategies used to overcome it.